Work

AI ThinkTank

Gemini
Multi Agent
LangChain
Multimodal

My submission to the Gemini AI Hackathon. A multimodal multi agent system that uses Gemini to simulate expert debates based on provided images.

Landing page of AI ThinkTank

I built this project for the Gemini AI Hackathon.

AI ThinkTank leverages the power of Large Language Models to simulate expert debates on any topic. Users can insert an image URL and the system automatically generates a topic of debate based on the image, using Gemini Pro Vision. Thereafter the system dynamically generates relevant experts who autonomously engage in a debate. Additionally, users have the option to join the discussion, adding an interactive element to the experience. AI ThinkTank was inspired by the paper on the MAD Framework to fight degeneration of thought and help people to open their mind to new viewpoints.

The system is built on top of the Streamlit framework, and uses the LangChain library to interact with Google’s Gemini Model.